Why 6G Intelligence still need to be explored

One of the main visions of 5G is edge computing, sometimes called mobile edge computing (MEC). The idea is that mobile network operators (Mnos) can deploy computing resources closer to base stations, enabling ultra-low latency and enabling Mnos to charge more for premium service.

Five years later, Mnos have barely deployed edge computing. Only a few operators have tried to partner with search engine companies to send data to those companies' cloud centers, rather than using their own computer resources.

Instead, the dominant model remains that telecommunications networks transmit data to "peers" (often referred to as Internet exchange points), where multiple Internet service providers and networks connect to exchange traffic. From there, the data is transferred to hyperscale-owned data centers for processing, and the responses are returned via the same path.

Why hasn't the edge computing concept caught on?

Edge computing promises to reduce latency by reducing the distance data processing must travel. However, fiber-optic cables transmit data at a speed of about 200 km/ms. A data center 100 kilometers away only adds 1 millisecond to the response time.

Current 5G networks typically have latency of 30-40 milliseconds, with the best private networks having latency of around 10 milliseconds. The data processing itself typically takes a few milliseconds, especially when it comes to video compression. Reducing response time by 1 millisecond by moving the calculation closer to the device makes little sense.

In addition, in most regions, there may be a large city within 1 millisecond that already has a hyperscale data center. These large facilities offer better economies of scale, and hyperscale businesses are better at selling computing services than Mnos. As a result, today's "edge" solutions largely mirror the traditional model, where Mnos transmit traffic to peers and then process it in the data centers of hyperscale enterprises.

One exception is private 5G networks, which route traffic to the IT systems of the private network owners. While this technically qualifies as edge computing, functionally IT's just another form of peering, this time into the local IT network.

Recently, the discussion about MEC has quieted down as Mnos have realized that edge computing is neither a viable deployment service nor a compelling revenue opportunity. In fact, Mnos are increasingly centralizing their own computing needs, consolidating baseband processing from multiple base stations into a centralized unit, rather than deploying computing resources from the edge of the network.

The future of edge Computing: Will 6G change anything?

For now, what 6G will look like remains uncertain. Mnos advocate "software-only" updates to reduce operating costs, while manufacturers are promoting "ultra-5G" with faster speeds and lower latency. Concepts such as perception and AI-native capabilities are also being discussed, but whether new spectrum will be allocated remains unclear.

For edge computing to take off, several conditions need to be met:

New applications requiring latency less than 5 milliseconds
Willing to pay more for this ultra-low latency service
Additional spectrum is allocated to support low latency air interfaces
6G deployments are broad enough to make edge computing viable across the region
None of this seems likely at the moment.

Most of the proposed 6G applications are simply reiterations of 5G promises, many of which have yet to be realized. Consumers and businesses have little interest in paying extra for 5G services, and securing extra spectrum for 6G is becoming increasingly difficult. In fact, by geographic location, mid-band 5G (3.5GHz) is only deployed on 20% of most countries, suggesting that 6G coverage will be more limited.

Will artificial intelligence and sensing technology be a game changer?

Two new applications that 6G often discusses are sensing and artificial intelligence. The market demand for sensing applications is still unclear, and implementing sensing in 6G May require high-frequency spectrum that is less suitable for communication.

Ai applications often require fast response times, which can justify edge computing. However, most AI workloads either run directly on mobile devices (e.g., AI assistants, visual processing) or require high-performance processing that is processed in large data centers. Few AI applications seem to require a millisecond response, and there is no strong market demand to pay for an extra millisecond of acceleration via edge computing.

Some industry leaders have suggested that Mnos sell their "idle" computing resources for AI workloads, leveraging unused capacity in their network baseband processing, but this thinking is flawed. Hyperscale data centers also have excess capacity during off-peak hours, making MNO computing resources unnecessary. In addition, the complexity of dynamically reallocating computing power between AI workloads and radio network functions is significant.

As a result, moving AI workloads to Mnos has little added value and is unlikely to be successful.

The edge still belongs to the cloud

The telecom industry has long tried to compete with hyperscale providers by offering value-added services, but has often failed. Mnos have struggled to gain traction in areas dominated by hyperscale providers and OTT providers, and edge computing is no exception.

In recent years, it has been shown that the true "edge" will continue to be in hyperscale data centers located in major urban hubs, rather than at the edge of the network.